How to Build Claude Skills to Automate Your Business

TL;DR

Build your bedrock skills first: a source-of-truth file (Atlas), a voice file (Echo), and a skill-builder (Genesis). Every other skill reads from these, so building out of order wastes days. With the foundation in place, later skills like post generators and prompt enhancers produce output in your actual voice automatically, and most take under an hour to build with no code.

Transcript

20 claude skills run roughly 80% of the operational work in my business today. They design carousels. They validate video ideas. They review contracts. They pressure test my biggest decisions. They run on a schedule whether I'm there or not. And most of them took less than an hour to build. None of them required a single line of code. And if I had ... Read More

Key Insights

• The stack is organized into five layers and 20 skills, where each layer feeds the next: the bedrock (source of truth), force multipliers (meta skills), the brain trust (pressure testing), the production floor (asset creation), and the ground crew (operational backbone).
• Build order matters because building out of order can waste days, while building in the right order lets the later skills practically build themselves once the bedrock exists underneath them.
• Atlas is the company's single source-of-truth file that every other skill reads first, holding the mission statement, ideal customer profiles, products and pricing tiers, the quarter's bets, and the things ruled out and why.
• Atlas is structured with addressable sections so each skill pulls only what it needs: the post writer pulls the audience section, Red Line pulls licensing, and Anchor pulls pricing, so no skill loads context it does not need.
• Echo is the voice file holding tone rules and examples that pull every Claude output back to a specific writing style, using cadence patterns, sentence length ratios, and words to use or avoid so output does not sound like a generic AI blog.
• Voice files work best with side-by-side examples because the model learns faster from examples than from rules; ten labeled real samples beat ten paragraphs of voice rules every time.
• Genesis is the skill that builds skills by interviewing you with questions until it has enough to write the skill markdown file, and it can also output sub-agent prompts, automation specs, or standalone SOP documents.
• Force-multiplier skills sharpen every other skill: Wetstone rewrites rough prompts into tight executable versions before a task runs, Probe interviews and pressure-tests automation requirements, and Blood Hound searches existing skill sources before you build anything new.

Questions & Answers

Q: What are the five layers of the Claude skills stack?

The stack has five layers holding 20 skills, where each layer feeds the next. Layer one is the bedrock, three skills that hold the truth of the entire business. Layer two is the force multipliers, five meta skills that make every other skill sharper. Layer three is the brain trust, three skills that pressure test high-stakes decisions. Layer four is the production floor, five skills that turn ideas into visible assets. Layer five is the ground crew, four skills that hold the operational backbone.

Q: Why should you build the bedrock skills first?

Building the bedrock first matters because skills above it are only as good as the foundation they read from. The creator learned this expensively by building a LinkedIn post generator before Echo, the voice skill. With the same model and prompt, the output sounded like a generic AI blog every time. After adding Echo and rebuilding the generator on top of it, the same prompts produced posts in the creator's actual voice. Building out of order can waste days, while the right order lets later skills build themselves.

Q: What is the Atlas skill and what does it contain?

Atlas is the company's source-of-truth file, a single document every other skill reads first so Claude never has to ask who you are. It contains the mission statement, ideal customer profiles with the exact problems each is trying to solve, every product and service with its pricing tier, the two bets being made this quarter, and the four things ruled out and why. It is built with addressable sections so each skill pulls only what it needs. The advice is to write it like onboarding a chief of staff who is smart but knows nothing about your company.

Q: How does the Echo voice skill improve Claude's output?

Echo is the voice file holding the tone, rules, and examples that pull every Claude output back to a specific writing style. It captures cadence patterns, sentence length ratios, words the creator uses, words to avoid, and real samples from past writing for each content type. Without Echo, Claude might write a long, hedged sentence like recognizing the limiting factor is the operational pipeline; with Echo it writes something short and clear like the bottleneck is not ideas, it is the pipeline. The key addition is side-by-side labeled examples, since the model learns faster from examples than from rules.

Q: What does the Genesis skill do?

Genesis is the skill that builds skills. You describe what you want, it interviews you with questions, and it writes the skill markdown file for you with no code required. For example, asking for a cold outreach email skill triggers Genesis to ask about the recipient avatar, the offer, the email length, the desired action, and past examples that worked, sometimes across ten back-and-forth exchanges. The creator's version also handles three other output formats besides skills: sub-agent prompts, automation specs for tools like Make or Zapier, and standalone SOP documents, because not every workflow needs to be a skill.

Q: How does the Wetstone prompt enhancer work?

Wetstone is a prompt enhancer that intercepts a rough or long prompt and returns a complete, tighter, executable version before the actual task runs. A vague input like write me a LinkedIn about AI agents would otherwise force Claude to guess the length, format, audience, and tone. Wetstone rewrites it into something specific, such as a 200-word LinkedIn post in the creator's voice for the AI founders avatar using a defined SOP structure with a hook, belief reframe, and call to action. It includes a constraint library that knows the creator's work, voice references, audience markers, and length targets, so it tightens based on the actual situation.

Q: What makes the Probe automation builder different from other interview skills?

Probe is an automation builder interview skill that takes any workflow and asks the right questions until the requirements are clean, then picks the best output format such as a skill, a routine, a scheduled task, a Notion database with an agent, or a manual SOP doc. What sets it apart is that it pressure tests its own recommendation by generating three failure modes, for example asking what happens if a guest social profile is private, if you cannot find an email, or if their content is in another language. You answer each one, making the requirements bulletproof before any skill gets written, whereas most interview skills only gather requirements.

Q: How does the Blood Hound skill help you avoid rebuilding skills?

Blood Hound is a skill search engine that checks every public and private skill source before you build anything, so you do not reinvent something that already exists. You type a job description, such as a skill that summarizes long meeting transcripts into action items, and it searches across sources like Anthropic's public skills repository, the co-work plugin repository, the MCP registry, your custom skills folder, and community marketplaces. It returns a ranked list with a build-or-borrow recommendation. The creator's custom version ranks results by source activity rather than only keyword match, so a skill in Anthropic's repo ranks higher than one in a community marketplace.

Summary & Key Takeaways

• The creator runs roughly 80% of business operational work through 20 Claude skills, each given a code name like Atlas, Echo, and Genesis. Most took under an hour to build and required no code. The skills design carousels, validate video ideas, review contracts, and pressure test decisions, running on a schedule whether the creator is present or not.

• The core principle is to build the bedrock first. Atlas holds the company truth, Echo holds the voice, and Genesis builds new skills. Skills above the bedrock are only as good as the foundation they sit on, proven when a LinkedIn generator built before Echo produced generic output but sounded authentic once rebuilt on top of Echo.

• The force multipliers make every other skill sharper. Wetstone intercepts rough prompts and returns tighter versions using a constraint library of voice and audience references. Probe interviews workflows and pressure tests requirements by generating failure modes. Blood Hound searches public and private skill sources, ranking results by source activity so you never rebuild what exists.